Reinforcement interval type-2 fuzzy controller design by online rule generation and q-value-aided ant colony...

Area of Science:

• Control Systems Engineering
• Artificial Intelligence
• Computational Intelligence

Background:

• Interval type-2 fuzzy systems (IT2FS) offer improved robustness to noise compared to type-1 fuzzy systems.
• Designing IT2FS, particularly the generation of fuzzy rules and tuning of parameters, remains a challenge.
• Existing reinforcement learning methods may not concurrently optimize IT2FS structure and parameters effectively.

Purpose of the Study:

• To propose a novel reinforcement learning method, online rule generation and Q-value-aided ant colony optimization (ORGQACO), for designing IT2FS.
• To concurrently design the structure and parameters of an IT2FS using ORGQACO.
• To demonstrate the effectiveness and noise robustness of the proposed ORGQACO method in control applications.

Main Methods:

• Development of an online interval type-2 rule generation method for evolving IT2FS structure and input space partitioning.
• Utilization of Q-values and a reinforcement local-global ant colony optimization algorithm for designing consequent part parameters.
• Integration of ant pheromone trails and Q-values, updated via reinforcement signals, for selecting consequent actions.

Main Results:

• The ORGQACO method successfully designed IT2FS for truck-backing, magnetic-levitation, and chaotic-system control problems.
• Comparative analysis showed ORGQACO's efficiency and effectiveness against other reinforcement learning methods.
• IT2FS designed with ORGQACO demonstrated superior noise robustness compared to type-1 fuzzy systems.

Conclusions:

• The proposed ORGQACO method provides an effective approach for the concurrent design of IT2FS structure and parameters.
• The use of interval type-2 fuzzy sets significantly enhances controller robustness to noise.
• ORGQACO is a promising reinforcement learning technique for advanced fuzzy controller design.